The top-left card shows the AI's task type and the severity badge (when a side effect is flagged). The top-right card holds the category pills you can correct. Check both before sending — if a category is wrong, click the correct pill and Save.

Side Effect: Severe means escalate to a provider immediately. Side Effect: Moderate means gather more information and reassess. Side Effect: Mild means an education response is appropriate. The badge only appears when a side effect is detected.

Appears when a message has both clinical and non-clinical parts. The Internal Note for Staff describes what the support team needs to do (ship a replacement, process a refund, transfer a pharmacy, etc.). You — whoever ran this triage — stay responsible for the patient reply. Share the internal note with the support team through whatever handoff method your team uses (a thread comment, an internal email, a ticket); don't include it in the message you send the patient.

Routine means respond in normal queue order. Same Day means the patient needs a reply today. URGENT means escalate now. If wrong, adjust the dropdown and click the green checkmark to save.

Clinical: Side Effect, Medication Question, Symptom Report, Dosing, Lab/Result Question, and similar. Non-clinical: Billing, Shipping, Account, Appointment, and General Inquiry. Non-clinical staff can only pick from the non-clinical list when correcting.

Any time the patient message is a reply to an ongoing conversation. Without the earlier turns the AI is triaging a fragment, and may re-explain things the patient already knows or re-ask questions they already answered. Click Add Prior Context to open the turn list. Each row has a speaker dropdown (Patient / Nurse / Other) and a text field; click + Add turn for each new turn. Put the oldest turn at the top and the most recent reply at the bottom. Then type the newest patient message in the main box below.

No. For a patient's first message — one that isn't a reply to anything — leave Prior Context empty.

Open your profile (the chip in the top-right corner) and click Inquiry History. It lists the inquiries processed through Relai. It doesn't auto-load — click Load when you want to see records.

Relai uses magic-link auth — no passwords. Enter your email on the sign-in page, click the link in the email you receive, and you're in. Your session stays valid in this browser until you click Sign Out or the session expires (rare; the app silently refreshes it in the background).

Go back to the sign-in page, enter your email, and request a new magic link. If you never receive it, check your spam folder and confirm your email is exactly what your admin used to invite you. If you're new and have never received an invite, ask your admin to create your account.

Open your profile (top-right chip) and click Sign Out at the bottom. Always sign out on shared computers — magic-link auth means anyone who can read your inbox could sign back in, but the session is local to the browser otherwise.

If the AI's classification or draft is clearly off, do not send it. Edit or rewrite the response, then use Submit & Learn with what you actually sent. If the category is wrong, correct it via the pill. Never copy a malformed or nonsensical draft to a patient — when in doubt, write the reply yourself.

After sending your response, if you changed the draft, paste what you actually sent and click Submit & Learn. The AI compares the two versions and saves a learning note. If you sent the draft verbatim, skip this entirely.

Click the correct category pill in the Classification card (Clinical or Non-Clinical), then click Save. Each correction is stored and builds a dataset that improves the AI's classification accuracy over time.

Go to Inquiry History, find the row in question, and click the × on the right side to remove the entire entry. Per-correction edits aren’t supported — the whole record goes, or it stays.

Use them to flag whether the draft was usable as-is. Thumbs up = good response, send the draft. Thumbs down = needs work — followed by a correction. Both feed into the dataset.

Everything the AI reads before responding to a patient message — protocols, side effect guidance, routing rules, urgency thresholds, and response templates. Any change takes effect immediately on the next triage.

Side Effects — education and management guidance.
Templates — response frameworks for common messages.
Protocols — if/then decision logic.
Rules & Notes — urgency thresholds, severity rules. Edit carefully.
Routing — when and how to escalate or route.
URLs — links you reference in patient responses.

Clinical staff (RNs/MDs) handle anything — clinical questions, non-clinical questions, escalations from non-clinical. Non-clinical staff (CSRs) handle non-clinical inquiries (Billing, Shipping, Account, etc.) and General Inquiry. When they receive clinical content, they hit Escalate to Clinical and the nursing team picks it up. Severity (mild/moderate/severe), clinical category selection, and editing clinical drafts are all clinical-only.

Paste the patient message and click Analyze. If the AI flags clinical content, you'll see a handoff card with the patient message and a generic acknowledgment ("I've passed your message to our nursing team..."). Copy the acknowledgment, send it to the patient through your usual channel, then click Mark as Escalated. The nursing team sees the inquiry in their queue and replies directly.

Non-clinical staff can edit non-clinical categories (Billing, Shipping, etc.) but cannot touch clinical categories or severity. Severity isn't editable from the UI by anyone — it's set by the AI. If non-clinical sees clinical content the AI missed, they hit Escalate to Clinical rather than picking a clinical category themselves.

Triaging a reply without Prior Context — the AI gets a fragment and produces a generic response.

Ignoring the red severity badge — Severe means escalate to a provider now.

Ignoring the routing card — if a non-clinical item is flagged, the internal note needs to reach the support team via your usual handoff method. Don't paste it into the patient-facing reply.

Urgency — Routine / Same Day / URGENT. How fast the patient needs a reply.

Severity — Mild / Moderate / Severe. How serious a flagged side effect is. Only shown when a side effect is detected.

Priority score — A 1–10 number combining severity, clinical content, and urgency. Higher = more pressing.

Escalation — Routing a message to clinical when non-clinical staff received it (or when the AI detected clinical content above non-clinical's scope).

Handoff template — The standardized acknowledgment a non-clinical staff member sends the patient when escalating clinical content.